The New Claude Opus 4.7 Feature Developers Are Obsessed With

Summary
In this video, presenter Mervin Praison reviews the release of Anthropic's Claude Opus 4.7, walking through its benchmark scores, features, and developer reactions. He details the model's new effort parameter levels, pricing, performance compared to earlier models and Claude Mythos Preview, and highlights developer features in Claude Code such as /ultrareview and auto mode.

What is shown

• [00:00] Overview of the Claude Opus 4.7 announcement post (dated 16 Apr 2026) and initial benchmark comparison table against Opus 4.6, GPT-5.4, Gemini 3.1 Pro, and Mythos Preview.
• [00:10] Review of highlight points from the announcement: instruction following, high-resolution multimodal support (up to 2576 pixels on the long edge), real-world knowledge work, and file-system memory.
• [00:44] GDPVal-AA knowledge work Elo score comparison chart (Opus 4.7 leading at 1753).
• [00:48] "Agentic coding performance by effort level" graph, illustrating performance versus token usage across low, medium, high, xhigh, and max settings.
• [01:06] Anthropic Python SDK code snippet showing how to set the output_config={"effort": "medium"} parameter.
• [01:36] Benchmark table showing Claude Mythos Preview outperforming Opus 4.7 on agentic coding and reasoning benchmarks.
• [01:43] API pricing and availability details across Claude products, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry.
• [01:54] Industry quotes and testimonials from Intuit and Augment Code.
• [02:19] Claude Code features explained: the /ultrareview command and the new auto mode security classifier system compared to --dangerously-skip-permissions.
• [02:54] 2D matrix diagram comparing task autonomy versus security/safety for manual prompts, bypass permissions, sandboxing, and auto mode.
• [03:08] Community discussions and charts on X (Twitter): Nathan Lambert on the new tokenizer/base model, MRCR v2 long-context benchmark degradation chart, Alex Albert's feature summary, and partner integrations/promotions on Cursor and Windsurf.

Claims & numbers

• Benchmarks & Scores:
  • SWE-bench Pro: Opus 4.7 scores 64.3% vs. Opus 4.6 (53.4%), GPT-5.4 (57.7%), Gemini 3.1 Pro (54.2%), and Mythos Preview (77.8%).
  • SWE-bench Verified: Opus 4.7 scores 87.6% vs. Opus 4.6 (80.8%), Gemini 3.1 Pro (80.6%), and Mythos Preview (93.9%).
  • Terminal-Bench 2.0: Opus 4.7 scores 69.4% vs. Opus 4.6 (65.4%), GPT-5.4 (75.1% self-reported), Gemini 3.1 Pro (68.5%), and Mythos Preview (82.0%).
  • Humanity's Last Exam (with tools): Opus 4.7 scores 54.7% vs. Opus 4.6 (53.3%), GPT-5.4 (54.7%), Gemini 3.1 Pro (51.4%), and Mythos Preview (64.7%).
  • GDPVal-AA Elo score: Opus 4.7 achieves 1753 vs. Opus 4.6 (1619), GPT-5.4 (1674), and Gemini 3.1 Pro (1314).
  • MRCR v2 (8-needle @ 1M context): Opus 4.7 drops to 32.2% (with thinking/max) compared to Opus 4.6 at 78.3% (64k thinking).
• Pricing & Parameters:
  • Pricing is unchanged from Opus 4.6: $5 per million input tokens, $25 per million output tokens.
  • Image input support increased to 2,576 pixels on the long edge (~3.75 megapixels), more than 3x prior Claude models.
  • Five effort tiers are available for Opus 4.7: low, medium, high, xhigh (new), and max.
  • Context window: 1M tokens.

Notable quotes

• [00:39] "I personally always use Claude Opus 4.6 for my coding purpose, but now we got 4.7."
• [01:01] "xhigh introduced only in Opus 4.7."
• [02:23] "The new /ultrareview slash command produces a dedicated review session that reads through changes and flags bugs and design issues..."

Assessment
This is an independent community commentary and overview video reviewing Anthropic's official blog posts, documentation, benchmark charts, and developer community reactions on X. The presenter does not run original benchmark evaluations or live code execution in the video, relying entirely on published tables, promotional blog posts, and third-party announcements.

Described by gemini-3.8-flash on 2026-09-29 from the video's audio and frames.